Regional and Sectoral Economic Studies Vol (2012)

Transcription

1 WAGE I EQUALITIES A D AGGLOMERATIO I CROATIA: WHO S TO BLAME? Vinko MUSTRA Blanka ŠKRABIC Paško BURNAĆ Abstract This paper uses NUTS sub-regional data to investigate the sources of the regional wages differences on the sector level from year forward in less developed country such as Croatia. The shift-share analysis is used to decompose the spatial variation of wages into productivy effect and occupational composion effect exploring how the spatial variance in wages is attributable to variations in these effects among different sectors. In the last step of the paper we examine the relationship between these two determinants of spatial differences and the agglomeration instrumented by measure of urbanization in the NUTS sub-regions. Keywords: sector composion index, sector-specific agglomeration, Croatia JEL classification: R, J. Introduction Regional inequalies are persistent feature of many economies, especially in less developed countries. One of the most important indicators of the inequalies is regional wage level. It is widely accepted that spatial effects have a different impact among sectors in developed countries. The tradional core-periphery pattern of manufacturing is weakening which tends to work towards regional convergence. On the other hand, service industries continue to concentrate in high-densy areas thus reinforcing divergence. Compared to developed, less developed countries have different economic path. Thus, is reasonable to test the influence of spatial effects among sectors for a less developed country such as Croatia. The starting point of the study is well documented growing disparies in wages across the Croatian NUTS III regions for period (Table.) Thus in maximum average wage (cy of Zagreb) was among Croatian NUTS III regions 9, % higher then the lowest average wage (county of Međimurska). Taking into consideration that the ratio was increased by 9,8 % from till is easily to conclude that growing disparies in wages are significant source of considerable policy concern. To explain large spatial wage disparies two broad sets of explanations can be proposed (Estaban, 999). First, differences in wages across areas could directly reflect spatial Vinko Muštra, Faculty of Economics, Universy of Spl, Matice hrvatske, Spl, Croatia, vmustra@efst.hr; Blanka Škrabić, blanka.skrabic@efst.hr; Paško Burnać, pasko.burnac@efst.hr

2 differences in productivy of the workforce. Thus, can be that in this region productivy per worker is above the mean of Croatia. Table. Average monthly wages (in kunas), UTS III regions, period MIN (Međimurska) MAX (Cy of Zagreb) Source: Central bureau of statistics- Republic of Croatia - Next explanation for wage differences is the occupational composion of employment. It implies that activies are not evenly distributed across areas and require different labor mixes. Consequently, we expect a higher mean wage in areas specialized in more skillintensive activies. (Rice et al.;.) For instance, the average productivy in agriculture, in industry or in the service sector could be identical across the regions. Yet, the regions specialized in services would have an average wage per worker higher than those specialized in agriculture. The present article extends the lerature in the following direction. After disintegrating the economy of Croatia in three (four) groups of sectors; (agriculture) manufacturing sector, market services and non-market services and decomposing the average earnings of each group on NUTS III region (county) level into a productivy effect and an occupational composion effect, article concentrates on occupational composion index for different sectors (sector composion index) or to be more precisely on relationship between these indexes and agglomeration instrumented by specific measure of urbanization in the NUTS sub-regions in Croatia appreciating different determinants of each effect and therefore different implications for policy makers. The paper is organized as follows. The next section covers data and descriptive issues, and derives data for decomposion of earnings into occupational composion effect and productivy effect for different sectors (agriculture, manufacturing, market services and For clarification is relevant to indicate that occupation stands for activy in National classification of activies. Main reason for maintaining occupation is methodological compatibily wh relevant papers. 8

3 Mustra,V., Skrabic,B., Burnac,P. Wage inequalies and Agglomeration in Croatia: Who s to blame? non-market services). The brief history of analysis of regional wage differences regarding shift-share analysis is reported in Section. Section concentrates on relationship between sector composion index for different group of sectors and economic mass presented wh a measure of urbanization, concentrating especially on non-market sector. Section presents conclusion wh directions for improving efficiency of regional policy.. Data and variables construction In the paper has been used panel data for the subregional NUTS spatial uns of Croatia, for the period, provided by Central Bureau of Statistics. The main advantages of using panel data are capturing both cross-section and time-series variation as well as allowing for meaningful inference using a sample wh a relatively small number of cross-section observations over a short time period. Allowing for dynamics in the underlying process is relevant not only to infer on the persistence of the series but also to ensure that the estimates for other parameters are consistent. Exceptions are variables for specific measure of urbanization. They cover period. In Croatia, NUTS subregional un is the level of county. In the analysis we used data for all counties. Variables are disaggregated two times. Firstly into activies according to National classification of activies and then in second stage these activies are grouped into three (four) sectors: (agriculture), manufacturing, market services and nonmarket services. Following papers by Rice et. al. () and Estaban (999) we present mechanical decomposion for separating contributions to the spatial structure of average wages on occupational composion index (OCI), productivy index (PI) and allocative index (AI). Symbols w and p denote the wage and level of employment in occupation j ijt ijt and area i for a period t. The average wage per hour and average level of employment of occupation j for a period t in the economy as a whole (i.e. aggregating across all i) is given by w and p. Thus the following equalies present average wages per hour in Croatia for a period t as a period t wh w : w jt t jt jt j jt w t and average wage per hour for NUTS III region (county) i for = p w and... () Now is possible to define the three components of the regional deviation in wages. The occupational composion index represent by OCI measures the differential wages accruing from region i's specific occupation composion in period t, assuming that occupation productivies in each region are equal to the Croatian averages for the same period. We thus wre: The main advantages of using panel data are capturing both cross-section and time-series variation as well as allowing for meaningful inference using a sample wh a relatively small number of cross-section observations over a short time period. Allowing for dynamics in the underlying process is relevant not only to infer on the persistence of the series but also to ensure that the estimates for other parameters are consistent. 9

4 OCI = ( pijt p jt ) wjt () j OCI takes posive values if the region is specialized ( pijt > p jt ) in occupation wh high wages at the Croatian level and de-specialized ( pijt < p jt ) in occupation wh low wage levels. OCI is at the maximum if the region is specialized in the occupation wh the highest wage in Croatia. The productivy index component PI presents contribution of occupation productivy differences to the shift between regional and national average wages, on the assumption that the region's occupational composion coincides wh national one. We then define PI as: PI = p jt ( wijt wjt ) () j PI is posive if the region has occupation productivies above the Croatian average. The allocative component AC is defined as: AC = ( pijt p jt )( wijt wjt ) () j This component is an indicator of the efficiency of each region in allocating s resources over the different industrial occupations and can also be viewed as measuring the covariance between occupations specialization and productivy advantages. Regarding the aim of the paper, or to be more precisely the analysis of the relationship between these indexes on the level of sector and agglomeration instrumented by specific measure of urbanization in the NUTS sub-regions in Croatia, is necessary to group the abovementioned activies into three (four) sectors. Considering the data limations four different options for this procedure are introduced in Table in the Annex. Due to grouping activies into sectors is necessary to adopt the mechanical decomposion. The adjustment is introduced in following manner: The sector composion index represent by Sm CI measures the differential wages accruing from region i's specific sector (m) composion in period t, assuming that sectors productivies in each region are equal to the Croatian averages for the same period. We thus wre: n S CI = ( p p ) wjt () m ijt jt j For clarification is relevant to emphasize that each sector (m) is defined as sum of activies from to n where m stands for sectors (agriculture, manufacturing, market services, non-market services) and n for number of activies in each sector.

5 Mustra,V., Skrabic,B., Burnac,P. Wage inequalies and Agglomeration in Croatia: Who s to blame? Sm CI takes posive values if the region is specialized in sectors wh high wages at the Croatian level and de-specialized in sectors wh low wage levels. SmCI is at the maximum if the region is specialized in the sectors wh the highest wage in Croatia. The productivy sector index component Pm SI presents contribution of sectors productivy differences to the shift between regional and national average wages, on the assumption that the region's composion coincides wh national one. We then define P SI as: m n Pm SI = p jt ( wijt wjt ) () m j P SI is posive if the region has sectors productivies above the Croatian average. The allocative sector component Am SC is defined as: n Am SC = ( pijt p jt )( wi jt wjt ) () j Interpretation of the A m SC also should be readjusted according aforesaid manner.. Decomposon of regional wage In this section will be provide evidences for modeling the determinants of wage differences according shift share analysis. Theoretical foundations have been presented in the articles by Rice et all. () and by J. Estaban (999) and could be summarized by central equation (8): w = α + OCI + PI + AC + ε (8) where w stands for average wage per hour in region i for period t, OCI for occupational composion index, PI for productivy index, AC for allocation index all for county i and period t. Constant term is presented by α and error term by ε for county i and period t. The next step implies empirical confirmation of the equation (8). Evidence could be found in abovementioned articles. In paper by Rice et. al. productivy index and occupational index have been confirmed for NUTS sub-regional data for Great Brain. J. Estaban in his attempt to elucidate the extent to which existing interregional inequaly per worker whin the EU can be attributed to differences in the composion of activies, rather than to productivy gaps that are uniform across sectors has also provided evidences for shift share analysis for EU regions. Regarding that abovementioned papers have empirical confirmation on developed countires, is necessary to introduce addional empirical argument introduced in paper by Muštra et al. (9) where abovementioned methodology was tested for regions of Croatia. Thus, the

6 exing lerature supports significance of productivy effect and, for our paper more important, occupational effect for regional wage differences for developed and less developed countries.. Agglomeration economies and sector composion effect Taking into consideration previous confirmation of occupational index is possible now to focus on main contribution of the paper, empirical confirmation that sector composion index is influenced by agglomeration instrumented by share of urban population and that this influence is different among sectors. Starting point for the empirical confirmation is the model wh the structure: T S CI = α + X β + δshare _ Urban _ primacy + ε (9) m where T X presents control variable vector wh dimensions x k in county i for a period t and variable Share _ Urban _ primacy share of the population of the biggest urban place in county i for a period t stands as a instrument for agglomeration. Control variables where chosen based on paper wrten by Xubei () in which education and age have been found as a significant variables for regional wage differences. Thus for control variable education has been used share of employed people wh higher education (share_high_education) and for age average number of year of employed people (AVGyear) and squared average number of year of employed people (AVGyear). It is assumed that ε are IID(, σ ε ) ; identically and independently distributed error terms. Procedure stars wh testing equation (9) for a four different options of grouping activies from National classification of activies already mentioned in table. The first challenging dilemma is recognizing is the process static or dynamic. The results of Durbin-Watson test for estimated static model (Fixed effect model, Random Effects model) under equation (9) indicate dynamic nature of our process and therefore we used dynamic model. Thus we introduce modified equation which includes dynamic behavior of dependent variable. This dynamic dimension is characterized by the presence of lagged dependent variable among the regressors. S CI = α + δsci + X β+ δshare _ Urban _ primacy + e..() T m i, t That renders the OLS estimator biased and inconsistent even if ε are not correlated. As a result, a new method for estimation was introduced by Arellano and Bond (99). In the paper two step Arellano and Bond GMM estimator is used because one step estimation assumes the error terms to be independent and homoskedastic across counties and over time. Two step estimator relaxes the assumption of independence and homoscedasticy Before starting we will indicate that support for the assumption that is possible to use share of urban population in each county as proxy for level of economic mass is documented for Croatia in paper wrten by Nejašmić (.) Education beyond the secondary level The omted results are available by request

7 Mustra,V., Skrabic,B., Burnac,P. Wage inequalies and Agglomeration in Croatia: Who s to blame? by using the residuals obtained from the first step estimation to construct a consistent estimate of the variance-covariance matrix. Thus, when the error term ε is heteroskedastic the two step estimator is more efficient (Cole, Moshirian and Wu, 8). Before introducing results of two step Arellano-Bond dynamic panel estimator in the Table, should be stressed that results indicate empirical relevance of urban primacy only for SCI of non market service sector which could be interpreted as evidence for the main contribution of the paper, or more precisely, as a evidence for hypothesis that influence of agglomeration on sector composion index is different among different sectors. Thus, Table represents only coefficients for sector composion index of non-market service sector. Table. The results of two step Arellano-Bond dynamic panel estimator Variable Option Option Option Option SCI_nms,*** (,),9*** (,),*** (,),9*** (,) avg_age 8,9*** (,9),*** (,8) 8,9*** (,9),*** (,8) avg_age -,8*** (,) -,*** (,) -,8*** (,) -,*** (,) share_high_education,*** (,),*** (,),*** (,),*** (,) share_urban_primacy,9*** (,9),*** (,),9*** (,9),*** (,) cons,8 ***(,8),*** (,),8 ***(,8),*** (,) umber of observations umber of individuals Sargan test (p-value),,,, m-test (p-value),**,**,**,** m-test (p-value),89,,89, *, **, ***- indicate significance at %, % and % level the number in brackets are standard errors Correlation coefficients share_high_education avg_age share_urban_primacy share_high_education. avg_age.8. share_urban_primacy.8.8. Source: Calculation by authors The omted results are available by request

8 It should be stressed that diagnostic tests (Sargan test and m statistics) for estimated model in Table are satisfying at % confidence level and therefore proposed model is well specified. The correlation coefficients between each of the variables are reported in the lower part of the Table. The highest coefficient of correlation is.8 and indicates that we should not expected high risk of multicolineary problem between variables of our interest. The coefficients on control variables are all statistical significant and have expected sign in all four options. The results on main variable confirm empirical relevance of our measure of agglomeration (Share_Urban_Primacy) in explaining SCI differences on significance level of %. It should be stressed that the magnude of coefficients should be carefully observed taking into consideration empirical limation of procurable data. Hence we concentrate on the sign of coefficients rather than on coefficients magnude. Thus sign of coefficient of the urban primacy variable implies significant evidence for posive relationship between sector composion index and agglomeration economies in Croatia. So, taking into consideration that sources of agglomeration might be economy-wide, location specific or sector specific (Bottazzi et. al., 8) the confirmation that the relation between SCI and urban primacy holds only for non-market services indicates sector specific dimension of the agglomeration economy in less developed country such as Croatia. What is possible theoretical background for confirmed sector-specific dimension of agglomeration economy? Starting point for looking the answer could be argument that reasons for entry firms and construction of sector composion index are similar. Therefore is possible to use lerature that tries to explain the regional differences in entry rates as a foundation for sector composion index explanation. The lerature usually identifies three major categories of factors influencing spatial differences: local demand factors, the supply of founders and the policy environment (Keeble et al. 99; Johnson and Parker 99). The local demand factors reflect the market potential for the new firms. Are there sufficiently many potential customers in the region and can they afford to buy the good or service supplied by the entrants? The supply of founders perspective focuses on who the individuals starting new firms are. What other opportunies than to start a new firm, do they have and what knowledge do the individuals in particular regions possess? The policy environment reflects, for example, what kind of support from local authories, in terms of financial support and knowledge support, is available to an individual that is planning to start a new firm. Also includes not only the measures for private subjects, but also location composion of public administration. All abovementioned factors consider agglomeration as a factor influencing a firm s decision to locate in or ex from a particular region and therefore main mark for possible theoretical explanation for confirmed sector-specific dimension of agglomeration economy.

9 Mustra,V., Skrabic,B., Burnac,P. Wage inequalies and Agglomeration in Croatia: Who s to blame? The results of our paper, empirical relevance of urban primacy only for SCI of non market service sector, emphasize primarily the role of policy environment, especially public sector as a key component of non-market service sector. Regarding that main contribution of the paper is empirical confirmation of sector specific dimension of the agglomeration economy in less developed country such as Croatia, full theoretical explanation could and should be main object of the another study. In the end, should be stressed importance of the result for further examination of wageinequalies and agglomeration. Firstly, may help to clarify whether, below the persistence of interregional inequalies in aggregate productivy per worker, there has been a convergence in composion sector by sector. Secondly, a crical role for the sector composion would cast doubts on the relevance of the aggregative one-sector growth models in explaining the regional differences. Finally, the results validate the appropriateness of the regional policy not only in Croatia but also in other countries wh experience of transion, essentially based on instruments geared to generate uniform productivy increases in backward regions (infrastructure and human capal).. Conclusion Growing disparies in wages are source of considerable policy concern. Therefore the paper tries to explore these inequalies by using shift share analysis. Shift share analysis presents mechanical decomposion of wages on occupational composion effect and productivy effect. It is widely accepted that spatial effects have a different impact among sectors in developed countries. On the other hand, less developed countries have different economic path. Thus, is reasonable to test the influence of spatial effects among sectors for a less developed country such as Croatia. Focus of the paper encompasses influence of economic mass for occupational composion for four different sectors (agriculture, manufacturing, market services, nonmarket services). For this purpose has been established new index, sector composion index, by aggregating similar activies from National classification of activies. The analysis of importance of the economic mass for sector composion index has been estimated by two step Arellano-Bond dynamic panel-data estimator and confirmed only for non-market service sector. In other words, the results indicate empirical relevance of urban primacy only for sector composion index of non market service sector which could be interpreted as evidence for hypothesis that influence of economic mass on sector composion index is different among different sectors. Taking into consideration that source of agglomeration might be economy-wide, location specific or sector specific the empirical confirmation of the relation between sector composion index and urban primacy only for no-market service sector indicates sector specific dimension of the agglomeration economy in less developed country such as Croatia. These findings should have significant effect on regional policy and therefore in next period research could be focused on recognizing and theoretical modeling relation between economic mass and sector composion appreciating possible different factors in developed countries and in less developed countries.

10 References Albouy, D., (9.), What are cies worth? Land rents, local productivy, and capalization of amenies values, National Bureau of Economic Research, Cambridge, Working paper 98 Arellano, M., Bond, S., (99.), Some Tests of Specification for Panel Data: Monte Carlo Evidence and a Application to Employment Equations, Review of Economic Studies, No. 8, Pp. -9 Barde, S.(8.) The spatial structure of French wages: Investigating the robustness of twostage least squares estimations of spatial autoregressive models, Documents de Travail, Observatoire Francais des Conjonctures Economiques (OFCE), No 8- Bottazzi, G., Dosi, G., Fagiolo, G., Secchi, A. (8.) Sectoral and geographical specificies in the spatial structure of economic activies, Structural Change and Economic Dynamics, Elsevier, No. 9, Pp 89- Ciccone, A. (.) Agglomeration effects in Europe, European Economic Review,, Pp. - Combes, P.P., Duranton, G., Gobillon, L. (), Spatial wage disparies: Sorting matters, Journal of Urban Economics, Vol., Issue, Pp -, March. Dunn, E.S. (9.), A statistical and analytical technique for regional analysis, Papers and proceedings of the Regional Science Association, No., Pp 9- Esteban, J. (999.), Regional convergence in Europe and the industry mix: a shift-share analysis, Regional Science and Urban Economics, Vol., Pp -,. Geppert, K., Stephan, A. (8.) Regional disparies in the European Union: Convergence and agglomeration, Papers in Regional science, Volume 8 (), Pp. 9-() Johnson, P., Parker, S. (99.) "Spatial Variations in the Determinants and Effects of Firm Births and Deaths," Regional Studies, Taylor and Francis Journals, vol. (), Pp. 9-88, November. Keeble, D., Walker, S., Robson, M., (99.), New firm formation and small business growth in Uned Kingdom: spatial and temporal variations and determinants, Research Series, No., Department of employment, London Muštra, V., Grčić, B., Burnać, P. (9): Sources of spatial differences in wages: Analysis for the regions of Croatia, Proceedings of the International Conference Economic Policy and Global Recession Vol., Pp. 9-8, Faculty of Economics, Universy of Belgrade Rice, P., Venables, A. J., Patacchini, E. (.) Spatial determinants of productivy: Analysis for the regions of Great Brain, Regional Science and Urban Economics, No, Pp - Rosenthal, S.S., Strange, W.C. (.) Evidence on the Nature and Sources of Agglomeration Economies, Handbook of Urban And Regional Economics, Volume,., Pp. 9- Xubei, L., (.), Regional disparies in labor market performance in Croatia: The role of individual and regional structural characteristics, World Bank Policy Research, Working paper, No. 8, March. Annex on line at the journal Webse:

11 Mustra,V., Skrabic,B., Burnac,P. Wage inequalies and Agglomeration in Croatia: Who s to blame? TABLES Table. Average monthly wages (in kunas), NUTS III regions, period... Table. Grouping activies from National classification of activies into sectors...8 Table. The results of two step Arellano-Bond dynamic panel estimator.... Count y yea r Empl_ total EMPL _ Man_ EMPL _ MS_ EMPL _ NMS_ WAGE _ TOT WAGE _ MAN_ WAGE _ MS_ WAGE _ NMS_

12 Table. county Ye ar avg_ wage perh_ tot avg_ wage perh_ man_ avg_ wage perh_ ms_ avg_ wage perh_ nms_ 8 avg_ age_ emp_ total share _urban share_ urban_ primacy share_ high_ educ

13 Mustra,V., Skrabic,B., Burnac,P. Wage inequalies and Agglomeration in Croatia: Who s to blame? Table. Grouping activies from ational classification of activies into sectors Actives two I.option II. Option III. Option IV. Option Agriculture, hunting and forestry Fishing Agriculture Agriculture Mining and quarrying Manufacturing Electricy, gas and water supply Manufacturing Manufacturing Manufacturing Manufacturing Construction Wholesale and retail trade; repair of motor vehicles, motorcycles and personal and household goods Hotels and restaurants Transport, storage and communication Market services Market services Market services Market services Financial intermediation 9

14 Real estate, renting and business activies Public administration and defence; compulsory social secury Education Health and social work Other communy, social and personal activies Private households wh employed persons Extra-terrorial organizations and bodies nonmarket service Source: Central Bureau of Statistics of Croatia and authors analysis Annex tables. Years - nonmarket service Market services nonmarket service nonmarket service nonmarket service Market services nonmarket service County year Empl_ total EMPL_ Man_ EMPL_ MS_ EMPL_ NMS_ WAGE_ TOT WAGE_ MAN_ WAGE_ MS_ WAGE_ NMS_

15 Mustra,V., Skrabic,B., Burnac,P. Wage inequalies and Agglomeration in Croatia: Who s to blame? County year Empl_ total WAGE_ TOT WAGE_ MAN_ WAGE_ MS_ WAGE_ NMS_ EMPL_ Man_ EMPL_ MS_ EMPL_ NMS_

16

17 Mustra,V., Skrabic,B., Burnac,P. Wage inequalies and Agglomeration in Croatia: Who s to blame?

18

19 Mustra,V., Skrabic,B., Burnac,P. Wage inequalies and Agglomeration in Croatia: Who s to blame? Table. Coun ty yea r WAGE perh_t OT WAGE perh_ma N_ WAGE perh_m S_ WAGE perh_nm S_ AG E_ EMP_TO TAL SHAR E _URB AN SHARE_URB AN_ PRIMACY SHARE_ HIGH_ED UC

20

21 Mustra,V., Skrabic,B., Burnac,P. Wage inequalies and Agglomeration in Croatia: Who s to blame?

22

23 Mustra,V., Skrabic,B., Burnac,P. Wage inequalies and Agglomeration in Croatia: Who s to blame?

24

25 Mustra,V., Skrabic,B., Burnac,P. Wage inequalies and Agglomeration in Croatia: Who s to blame?